核心信息
AI智能体在调试时能一次性列出所有可能原因，对代码插桩并执行一次即可定位真正问题，而人类通常只能逐个排查最明显的原因，反复尝试才能找到答案。
要点
- 人类调试是串行的：先猜最可能的原因，检查，再重复，直到找到问题。
- 智能体可以并行验证多个假设，只要环境合适，无需显式指示它也会这样做。
- 把大语言模型当成人来用会降低效率：人无法把一个任务拆成五份并非线性地并行推进，但智能体可以。
AI智能体调试新思路：一次性验证所有假设
原始推文
an example of this is bug fixing you typically would think of the most obvious cause and check. then repeat until you find it the agent can come up with all potential causes at once, instrument the code and execute it once to figure out which one caused it
引用推文
one side effect of treating LLMs like humans is people keep using them ineffectively you can't break a task up into 5 pieces and work on all of them in parallel and non-linearly but the agent can. you don't even have to tell it to, if it has the right environment it will do it